Inside the iPhone 6s

[astilo]

View attachment 15454

And finally here we are, thanks to the Chipworks team for their nice, as usual, job.
Inside the iPhone 6s | Chipworks

The APL0898 size fits with an 80% shrink on the A8, and seems to have 8 MB cache - and our first look leads us to believe that our sample is from Samsung

View attachment 15455

For the first time we have seen, Micron has put some memory into the iPhone (though iFixit’s sample has Samsung), and it’s up to 2 GB for the first time for an Apple phone. As expected, they are dual-sourcing (at least) the flash - we have Hynix, iFixit has Toshiba.

2 x Knowles KSM2 microphones
2 x Apple/Cirrus Logic 338S1285 Audio IC (likely an iteration of the 338S1202 audio codec found in the iPhone 5s)
Apple 343S00014 3D Touch Controller?
Apple A9 APL0898 application processor
Apple/Cirrus Logic 338S00105 Audio IC
Apple/Dialog 338S00120 Power Management IC
Avago ACPM 7714 Multimode Power Amplifier
Avago AFEM-8030 Power Amplifier Module
Bosch Sensortec 367 LA 3-axis Accelerometer (likely BMA280)
Bosch Sensortec barometric pressure sensor BMP280 ?
Goertek GWM1 microphone
InvenSense MP67B 6-axis Gyroscope and Accelerometer Combo (also found in iPhone 6)
Micron D9SND (MT53B256M64D2NL) 2 GB LPDDR4 SDRAM
Murata 240 Front-End Module
Murata(?) Ne G98 RF Front-End Module
Murata(?) Yd G54 RF Front-End Module
NXP 1610A3 (likely an iteration of the 1610A1 found in the iPhone 5s and 5c)
NXP 66V10 NFC Controller (vs. 65V10 found in iPhone 6)
Qorvo/RFMD RF1347 Antenna Switch Module
Qorvo/TriQuint TQF6405 Power Amplifier Module
Qualcomm MDM9635M LTE Cat. 6 Modem (vs. the MDM9625M found in the iPhone 6)
Qualcomm PMD9635 Power Management IC
Qualcomm QFE1100 Envelope Tracking IC
Qualcomm WTR3925 Radio Frequency Transceiver
RF Micro Devices RF5150 Antenna Switch
SK Hynix H230DG8UD1ACS 16 GB NAND Flash
Skyworks SKY77357 Power Amplifier Module (likely an iteration of the SKY77354)
Skyworks SKY77812 Power Amplifier Module
Texas Instruments 3539 Retina display driver
Texas Instruments 65730AOP Power Management IC
Texas Instruments 6BB27
Texas Instruments SN2400AB0 Charger IC retained from the iPhone 6
Universal Scientific Industrial 339S00043 Wi-Fi Module

 

[astilo]

Some credits of course also to iFixit
[video=youtube;ROCzV9gMuA0]https://www.youtube.com/watch?t=8&v=ROCzV9gMuA0[/video]

 

[Tom Simon]

The board design technology is very impressive. They have smd devices arrayed across both sides of the board, with dense BGA chips on both sides as well. This board has many devices that be victims of EM noise and many emitters.

The other curious thing is that they have multiple accelerometers. Bosch and Invensense both have mems chips in this design. Are they just using the Invensense gyro and using the Bosch accelerometer for sensor fusion?

Also interesting are the 3 PMIC's from different sources.

 

[Daniel Nenni]

Pricing date from Forbes:

 

[Don Dingee]

No separate motion coprocessor - it is now integrated on the A9.

 

[nick_rb]

@ Tom

The 6-axis IMU has a higher precision than the 3-axis accelerometer, but a higher power consumption. As such, the IMU will be switched off as often as possible to save battery.

 

[Daniel Nenni]

My record of guessing where Apple will manufacture their Ax SOCs is officially broken:

Chipworks: Both Samsung and TSMC are making the A9 chip for Apple | Ars Technica

Not only do I owe a certain person a VERY expensive bottle of wine, I hang my head in shame. The good news is that we will get some never before available intelligence of identical SoCs on the same process in additional to different SoCs on the identical process (Exynos, SnapDragon 802, Apple A9, etc...).

Interesting times in the fabless semiconductor ecosystem for sure!

 

[lefty]

Well firstly, predicting the future is difficult. Secondly, you have the honesty to admit you were wrong - most people just stay quite and hope no one will notice. And thirdly, hats off to the people who got it right:
Login to DIGITIMES archive & research

 

[hist78]

My record of guessing where Apple will manufacture their Ax SOCs is officially broken:

Chipworks: Both Samsung and TSMC are making the A9 chip for Apple | Ars Technica

Not only do I owe a certain person a VERY expensive bottle of wine, I hang my head in shame. The good news is that we will get some never before available intelligence of identical SoCs on the same process in additional to different SoCs on the identical process (Exynos, SnapDragon 802, Apple A9, etc...).

Interesting times in the fabless semiconductor ecosystem for sure!

Daniel,

It's not that bad. Don't miss the next guessing game which is what's the percentage in terms of A9/A9X orders Apple awarded to Samsung and TSMC.

I think A9x will follow your original prediction to be 100% produced by TSMC. On the other hand I think TSMC will make at least 40% of all A9 while Samsung will do up to 60% of A9. The uncertainty for the order distribution is due to the current global economic situation and Apple's insistence to get further price concession from Samsung and TSMC. The rumor is TSMC prefers to maintain their sizable profit margin they feel reasonable and fair while Samsung caved in to Apple's price cutting demand.

Sorry, I don't owe any expensive wine to anyone so I can do this prediction freely.

 

[Speedfriend]

Any forecasts on how similar the performance will turn out to be in terms of battery life?
Also any guesses as to the additional cost to tape out and test for a second source?

 

[astilo]

While the split ratio between SAM and TSMC was unknown (and not yet disclosed), I was extremely confident about Apple doing double sourcing for the A9 SoC:
https://www.semiwiki.com/forum/f302/buying-opportunity-tsm-5690-3.html#post19637

@ Speedfriend
The cost figure is huge for sure, but part of the risk mitigation (and price negotiation) strategy of Apple. They plan to sell something like 80-90 million of Iphone 6s/6s+ during the first year. They can afford it. Moreover, eventually the foundries will pay for it, because of the direct competition.
Performance wise, it will be very hard to see noticeable differences driven by the 2 SoCs. The max CPU frequency has been for sure wisely chosen.

 

[lefty]

TSMC's 16 FinFET+ is ment to be faster than Samsung 14nm, so this means some iPhone 6s will be faster than others?

 

[astilo]

TSMC's 16 FinFET+ is ment to be faster than Samsung 14nm, so this means some iPhone 6s will be faster than others?

Not 100% sure about that (I mean, 16FF+ being faster than 14LPE). Anyway, speed is going to be pretty much the same, since the clock is the same (unless there is also a different behaviour in the thermal throttling of the SoC).
What could be different, is the battery drain, because xtors leakage might be different.
For sure density is different. Since Samsung and GF can put more chips on the same wafer, TSMC yield must be higher in order to be competitive in terms of price. We are in fact talking about something like 620 potential dice vs 570 potential yielding dice per wafer. Not negligible.

 

[Daniel Nenni]

Daniel,

It's not that bad. Don't miss the next guessing game which is what's the percentage in terms of A9/A9X orders Apple awarded to Samsung and TSMC.

I think A9x will follow your original prediction to be 100% produced by TSMC. On the other hand I think TSMC will make at least 40% of all A9 while Samsung will do up to 60% of A9. The uncertainty for the order distribution is due to the current global economic situation and Apple's insistence to get further price concession from Samsung and TSMC. The rumor is TSMC prefers to maintain their sizable profit margin they feel reasonable and fair while Samsung caved in to Apple's price cutting demand.

Sorry, I don't owe any expensive wine to anyone so I can do this prediction freely.

I had originally predicted a 70-30% split giving Samsung the majority based on sales of the iPhone versus iPad. Now I would say it is still a 70-30% with TSMC getting majority. I also predict that TSMC will get 100% of the iPhone/iPad business in 2016 which begs the question: Will Samsung stay in the foundry business without Apple?

 

[lefty]

Not 100% sure about that (I mean, 16FF+ being faster than 14LPE). Anyway, speed is going to be pretty much the same, since the clock is the same (unless there is also a different behaviour in the thermal throttling of the SoC).
What could be different, is the battery drain, because xtors leakage might be different.
For sure density is different. Since Samsung and GF can put more chips on the same wafer, TSMC yield must be higher in order to be competitive in terms of price. We are in fact talking about something like 620 potential dice vs 570 potential yielding dice per wafer. Not negligible.

TSMC has said many times that it is faster (In fact, they said that it's faster than 14LPP).. and of course, the clock speed varies with thermal limits, so you should see some difference.
I would hope that Chipworks would benchmark their samples. This is the first time we have an opportunity to directly compare performance of two competing nodes.

 

[Daniel Nenni]

Also any guesses as to the additional cost to tape out and test for a second source?

I'm looking into this but from what I'm told today TSMC did the migration for Apple. It was described to me as a "diving catch" to get more Apple business betting that Samsung would not meet Apple's yield-schedule-capacity requirements. Based on what I know about Morris Chang and how competitive he is this is definitely a strong possibility.

I'm in Silicon Valley today and this is what everyone is talking about. Mostly giving me grief for being wrong but also some interesting data points. If you have more info post it here or send me a private message. I will do a full analysis this weekend so stay tuned.

 

[hist78]

I had originally predicted a 70-30% split giving Samsung the majority based on sales of the iPhone versus iPad. Now I would say it is still a 70-30% with TSMC getting majority. I also predict that TSMC will get 100% of the iPhone/iPad business in 2016 which begs the question: Will Samsung stay in the foundry business without Apple?

In early August, we had a discussion regarding Samsung Semi 2Q2015 revenue. I mentioned that compare 2Q2105 and 2Q2014 Samsung Semi's revenue, I can't find a good reason to explain why Samsung's revenue growth is flat while "getting" a majority A9?

https://www.semiwiki.com/forum/f300/samsung-smartphone-declines-6337-post23349.html#post23349

We will know how close your prediction is once Samsung Semi release their 3Q2015 number.

 

[lefty]

So, what are the theories why Apple split the A9 order?
I think there are a couple of possible reasons:
1) Not enough supply. Last year Apple were not able to meet the demand for iPhones. So, given that a new node is not going to ramp up instantanously, they had to split the manufacture to guarantee enough SoCs for the launch.
2) They really wanted TSMC 16 FinFET+ because it performs better and has better yields, however TSMC was late, so they used Samsung as a stop-gap - otherwise they would have had to delay the launch.
3) a combination of the above?

 

[Brianhayes]

My record of guessing where Apple will manufacture their Ax SOCs is officially broken:

Chipworks: Both Samsung and TSMC are making the A9 chip for Apple | Ars Technica

Not only do I owe a certain person a VERY expensive bottle of wine, I hang my head in shame. The good news is that we will get some never before available intelligence of identical SoCs on the same process in additional to different SoCs on the identical process (Exynos, SnapDragon 802, Apple A9, etc...).

Interesting times in the fabless semiconductor ecosystem for sure!

Please un-hang your head in shame and continue your informed guessing. It is immensely valuable. Only a foolish reader expects every guess to be right every time,

 

[hist78]

So, what are the theories why Apple split the A9 order?
I think there are a couple of possible reasons:
1) Not enough supply. Last year Apple were not able to meet the demand for iPhones. So, given that a new node is not going to ramp up instantanously, they had to split the manufacture to guarantee enough SoCs for the launch.
2) They really wanted TSMC 16 FinFET+ because it performs better and has better yields, however TSMC was late, so they used Samsung as a stop-gap - otherwise they would have had to delay the launch.
3) a combination of the above?

Another possibility: Apple tries to gain every bit of price concession from Samsung and TSMC and use one against the other. On the way doing it, Apple also tries to hurt/drag down/tie down/slow down Samsung whenever it's possible.

Samsung knows Apple's game very well, but they have to play along due to the difficult market condition across all Samsung product divisions, from consumer electronics, smartphones, memory, to foundry. Samsung knows if they say "no" to Apple's demand, the worst case Apple will suffer is TSMC again becomes the sole source of A* SoC (probably not a big deal for Apple).

On the other hand, Apple has to maintain significant business and with TSMC to make this dual sourcing work even if Apple can't win all their pricing demand with TSMC. The reason is very simple: Apple doesn't want Samsung to be the single source for A* Soc, no more. Otherwise why Apple wants to go through all those lawsuits and agony against Samsung?